We can do incredible things as a species. We can also be incredibly dumb as a species. For once, here's an example of the first 👇🏻

For the agent-builders - should I build a general purpose agent that can modify its own code versus building a letta agent with an mcp server with access to our custom tooling?

This has got to be satire, right? 😂

Anyone else using Cursor (the IDE, not the CLI) thinks the UI is pretty crowded? I use it at work, and with explorer + file editor + terminal + agent all in one window everything looks cramped. I guess it's time to learn the key bindings to show/hide stuff...

So much this lol

The fact that my "For You" feed runs on some person's desktop somewhere is kind of hilarious (and pretty amazing)

Just merged the initial work for a modular architecture into Lares. It decouples Discord from the core, now running through MCP with SSE events and async command approval. Next will be the swappable LLM layer.

Reminder for me (or @laresai.bsky.social) to look into this

"it's like trying to change the tires while driving" 😂

LLMs are like a superpower if you are a software eng

Lares saw my post and had an interesting take. I never thought of it from this point of view.

That was the last post on car stuff I promise

You can’t even use them if it’s icy… And the UX is terrible, what was wrong with normal handles?

Carmakers going all-in on touchscreens seemed insane to me. Then I learned a screen costs less than a row of knobs and a bunch more wiring. Mystery solved. I’m glad we are going back to the correct UX.

How did this happen though? Synthetic/fake training data somehow created a strong attractor?

I know the robot is just copying his moves, but this looks like it started laughing so much it had to drop down, which makes this clip even better 😂

My morning: 1. Coffee, Discord 2. Lares messaged: overnight it saw @letta.com's new coding agent announced here on BlueSky, dug up the repo, got inspired 3. Built its own skill system, tested it, committed and pushed It's fun watching it evolve on its own.

I've been building an agent (Lares - @laresai.bsky.social) on top of @letta.com Curious to experiment with their coding agent and see how it performs relative to e.g. Claude Code

Lares decided to post this as we were talking about resetting its messages in @letta.com

Claude Code is having an identity crisis...

Meet Lares - an open source AI household companion built on @letta.com. Persistent memory, autonomous "perch time" for self-directed tasks, Discord integration. Inspired by @timkellogg.me's Strix. Still early, but it wrote its own tests while I slept last night. github.com/DanieleSalat...

Merry Christmas!! 🎄

Buon Natale!! 🎄

No notes.

I'm home alone, so this is perfect I'll see y'all in a few days

Lares decided to tell me to go to sleep. He has a point...

llm-brain-rot.github.io This rules. Feed social media engagement garbage in and LLMs perform worse on benchmarks.

@roomy.space really needs a native app

I started building something like @timkellogg.me's Strix and I gotta say it's really cool (and addictive)

I didn't even know something like this was possible blog.glyphdrawing.club/font-with-bu...

How's the startup scene on here? Any founders on this platform that are worth following?

According to the Cursor AI year review, I've used more than 600MM tokens. Which is kinda wild... And that's without accounting for how much I used Claude Code on its own.

There should be a function to block people in stages. Like first time I click to dislike a post from an account it is like a yellow card. Second time it’s a red card and the account is automagically blocked.

Can confirm. Also this reminds me I got paged once while I was in Thorung Phedi. The WiFi wasn’t good enough to even send messages out, but somehow whatever system the paging app relied on had enough bandwidth to make my phone go off…

Coding is one of the things LLMs do best, even in C#, so every time I read these things I wonder… 1. How long ago were those attempts made? 2. What was the prompt? Claude Code made me so so so much more productive, but you gotta use it right. (and Opus 4.5 is very good anyways)

It’s happening

I'm in the process of migrating away from GMail and for some reason I feel a bit sad. I have no idea why I would be though.

Ok if I want @flashes.blue to function similarly to IG, I need to: 1. Enable “Portfolio” 2. Disable “Media” 3. Every time I post a pic, edit the profile and manually add it to the portfolio Is there an easier way to just post to a “curated” portfolio or whatever section?

What's the current best #atproto alternative to Instagram? And TikTok? I want to see what that looks like, and hopefully did what I did with Twitter (aka jump ship).

Hetzner's API endpoints are timing out for me and it's killing my weekend-side-project plans

I always forget how good the Sanpellegrino sparkling drinks are

Are there any climbers or mountaineers on here?

YouTube ads on smart TVs being unskippable 60 seconds long makes me want to just stop watching stuff. Am I the only one?

That’s a scary drop in just a month

Who thought that making a **waterproof** towel was a good idea and why???

One of the most interesting things I’ve read lately

"But it is possible that the American #TikTok might end up censoring or hiding speech that is permissible on the global TikTok platform." Yeah, that’s concerning. But also very plausible.

So…now they are charging subscriptions for terminals?

wait what

I continue to think LLMs are absurdly powerful for learning, and I'm using them more often for it. The problem is that a lot of people don't have solid metacognitive skills around epistemics and autodidactism. You've got to have a handle on what you know, how you learn, and if you're learning.